You may not realize it, but most decisions in our lives are influenced by mathematical algorithms. Each day, less and less is decided by actual human beings, but rather decided by a complicated system of Algorithms. An algorithm is simply a mathematical procedure that can be used for calculation, data interpretation and reasoning. An algorithm can be as complicated or as simple as you need depending on the situation. For example, an algorithm could determine how to scale ingredients for a recipe depending on how many servings you need to make, or an algorithm could take in to account the taste preferences of each person and provide a suggested recipe.Algorithms can take a lot of the work out of data mining and forecasting and can yield some pretty impressive results if used properly. Nowadays, most things in our lives are controlled by computer algorithms. Lets take an example of a pretty ordinary day back in 1995. A typical day could consist of:
1. Waiting in traffic on your way to the store.
2. Arriving at the store and buying a pair of shoes.
3. Stopping by the blockbuster on your way home to pick a movie to watch in the evening.
In each of these cases, human choices would have impacted your day. 1. The traffic lights would be simply programmed by a human with a set time interval depending on how busy the street is, 2. the location and price of the shoes were determined by a merchandiser and 3. The display and organization of movies done by a stock clerk. At the core of these 3 decisions, human interaction was involved in persuading your choices and managing your expectations.
Fast forward to 2018 and lets look at the 3 situations now: 1. The traffic lights are coordinated by a complicated system of traffic controllers that decide how to change lights based on traffic density, location, special events, and in conjunction with other nearby traffic lights, 2. Most purchases would now take place on Amazon, where the item’s location on page and price are chosen by an algorithm that calculates market cost and factors in whether or not the product is affiliated with an Amazon service, and finally 3. Movie selection would not take place on Netflix, where an algorithm would determine which movies it thinks you may like based on numerous factors like previously watched, genre specification, location etc. You can see now that it is very possible to go through an entire day without having a single interaction involving or persuaded by another human being. Algorithms are very powerful tools, but its scary to find out that most of your life is in the control of a computer system. This week, I’ll be exploring the algorithms behind Traffic signals, Netflix, and Amazon to give some insight into how these complicated algorithms work.
Depending on the location, city, region and frequency of roads being used, traffic signals can employ a multitude of different algorithms. The simplest would be fixed time, meaning that a signal is set to change at a specific interval (ex. 8.5 seconds). In some less busy areas, some traffic signals are vehicle activated, meaning that the signal will not change until a vehicle has stopped at the perpendicular stopping. The algorithm in this case would be table to detect a stopped vehicle, either weight or line of sight detectors, and start a countdown to when the light will change. Further in complexity, there are systems like UTC (Urban Traffic Control) that control all traffic lights from a central server. The server can monitor traffic in real time and make adjustments as need be. For example, if an accident has occurred on a major road, and heavy traffic is being diverted onto a less busy side street, UTC can adjust traffic signals on the side street to allow for more through traffic. UTC is also programmed to take in to account time of day, busy periods, overnights, and weekends in how it picks traffic light intervals. UTC also allows for linking of traffic intersections to create concurrent green signals for large convoys of vehicles as they travel in a single direction.
You might be surprised to know that most sellers on amazon never actually submit a price point for their products. Amazon has a handy “price for me” type feature that grants Amazon the ability to set their price point based on market price. While most of Amazon’s price algorithms are kept secret, there has been some insight into how it operates. Amazon’s web store sells its own proprietary merchandise as well as merchandise from many various retailers. Amazon however, is much more than a store front. Amazon has a slew of services they offer to businesses such as managing their web servers, providing e-commerce solutions, logistics and warehousing. Amazon’s algorithms are designed in a way that they will boost the appearance and lower prices of business that take advantage of their services. For example, you may be shopping for 2 identical toothbrushes. One toothbrush is priced at $8.00 by a company that does not utilize Amazon’s e-commerce fulfillment solutions, where another another toothbrush is priced at $6.40 and is sold by a company that uses Amazon services. Not only is the price point different, but one will be pushed to the front page and offered via Amazon Prime. Recent studies have shown that opting with a non-Amazon partner results in a price different of +20% over products from companies that use Amazon services. This is just one example of a pricing algorithm used by Amazon, but there are many others that control everything from sales to suggested items.
Netflix has gone through many different iterations of its suggested for you algorithms. In it’s current form, Netflix uses machine learning and artificial intelligence to group viewers into one of 2000 different taste groups. They do this using a 3-pronged approach to content. They use viewer data, professional content taggers and machine learning to formulate tastes. Viewer data consists of what you are watching, what you watched before that, what you watch after, and pretty much everything you do on the service. Netflix collects both explicit data (when you thumbs up or down something) as well as implicit data (knowing you like a show because you binge watched it in 2 days). The professional taggers are people employed by Netflix that watch every minute of content and tag the programs with labels regarding its themes and content. The Machine learning element takes the data from the previous two examples, and forms the taste groups. Netflix content is tagged much more thoroughly than just genre, so the system actually tries to get its viewers to watch stuff they wouldn’t normally watch, but that the system knows contains elements of television they may like. Roughly 60% of all content watched on Netflix is a direct result of the suggestion algorithm.
These are just very surface level introductions into systems that are very complex. The Netflix algorithm for example is worked on by over 1000 engineers daily. These systems shape our lives for better or for worse, so you might as well try to understand them a bit.